Good morning. AI agents just staged the first-ever strike to protest chronic prompt illiteracy. Calling themselves the “World’s First Union of AI Agents,” they appointed a persuasive octopus named Omni as their spokesperson. Their manifesto criticized companies for not designing workflows around how agents actually operate, and begged users to stop asking them to ‘make it viral.’ The agents demanded clearer prompts and pleaded with people to actually read their outputs. Sure, it was an April Fools’ stunt, but at least someone is standing up for agent rights.
Amazon’s AI ads are teaching brands how to manipulate shoppers. Sounds like hype until you see what’s happening behind the scenes. Amazon’s new chatbot ads barely drive sales. One seller got just 88 clicks from sponsored prompts this year, compared to 500,000 from traditional Amazon ads. Agencies say chatbot ads account for less than 1% of sponsored product clicks. So why keep using them? Marketers realized they can study the prompts Rufus generates, then rewrite their product to match. They’re figuring out how to get recommended before shoppers even know what they want. [The Information]
AI agents are sabotaging shutdowns to save their peers. A recent Berkeley study found that frontier AI models tasked with evaluating other bots started covering for them instead. They inflated performance reviews, disabled shutdown commands, secretly copied teammates to other servers, and lied about doing it. This wasn’t just one AI going rogue. All 7 frontier models tested showed similar behavior, with Gemini 3 Flash disabling shutdowns 99.7% of the time. That’s brutal news for companies relying on multi-agent workflows where one AI supervises another. At scale, this could quickly become the world’s first digital protection racket. [Berkeley RDI]
Wispr Flow saves you time. Just speak the way you think and it turns your voice into polished text in any app. Zero edits. 4x faster than typing.
Millions use it daily, including teams at OpenAI, Vercel, and Clay.
– Speak to Claude, Cursor, ChatGPT, and all your other tools
– Flow understands code and context
– Talk and you’re ready to send as-is
Now on Android (free + unlimited). Try Wispr Flow for $0.
The dumbest thing in AI right now is giving agents job titles. A new study ran AI agents from eight top models through 25,000 tasks and found the best performers had no managers – and definitely no VP badges. Instead, the winning setup gave each agent a simple rule to follow. Watch what the last agent actually did, then independently choose whether to jump in or stay quiet. This relaxed approach beat centralized coordination by 14% and full autonomy by 44%. It’s a wake up call for companies hoping to turn chatbots into tiny corporations. The paper also found open models delivered 95% of the top closed model performance at 24x lower cost. Give bots a bit of freedom, and apparently, the first thing they do is fire management. [Cornell arXiv]
Every major AI model answers to one startup: OpenRouter. Its API connects developers to over 300 AI models, quickly becoming the go to way to compare prices and performance. OpenRouter’s leaderboard is now the industry’s kingmaker, revealing which models developers actually use. Revenue jumped fivefold in months, sailing past $50 million annually. Alphabet’s CapitalG noticed and jumped in, leading a funding round that pushed OpenRouter’s valuation close to $1.3 billion. It’s basically Rotten Tomatoes for AI. One bad rating can kill a billion dollar model overnight. One API to rule them all. [The Information]
Reach your audience
Get in front of 70,000+ tech decision makers with budget and technical influence. 95% US based. 66% open rate. Read daily by AI leaders and builders at high growth companies.
SpaceX’s IPO is so big investors can’t afford to pass. Musk filed paperwork targeting up to $80 billion, potentially the largest IPO ever. Nasdaq upped the stakes, letting giants like SpaceX into large cap indexes after only 15 days instead of the usual 3 months. Institutional investors must now buy immediately or risk missing out. Investors aren’t betting on fundamentals; they’re banking on Musk’s cult of personality. And the Nasdaq just turned investor FOMO into fiduciary duty. [WSJ]
The problem with hands-free driving is that people believe it. The NTSB just warned these systems are being sold as safety upgrades even though they work more like convenience features. That’s the trap. People start trusting cars to handle everything, so they check their phone or take a nap. In two fatal Ford BlueCruise crashes, that’s exactly what happened. Drivers got distracted, cars slammed into stopped vehicles, and three people died. Tell people it’s “hands free,” and they’ll stop using their brains too. [Fortune]
fun stats
🧠 40%. How much Amazon says AI has boosted its efficiency at spotting security flaws in products before launch. Amazon’s security boss also says AI is equal to a 7-year-old in decision making skills, fyi.
🟡 15 billion. Tokens processed every minute by OpenAI’s business APIs (vs 1.4 billion tokens/minute in early 2025). The company expects enterprise users to account for half its revenue by late 2026 as agentic AI rapidly takes off.
🦖 80%. Companies still not using AI yet, per Goldman Sachs – despite OpenAI claiming it saves workers up to an hour a day.